Fine-Tuned EfficientNetB4 Transfer Learning Model for Weather Classification

In meteorology and climate research, weather classification is known as weather classification, identifying the present weather patterns based on readings and observations of numerous environmental characteristics, such as temperature, humidity, pressure, and precipitation. This study presents a deep learning model for classifying the weather that uses the Adam optimizer and the EfficientNetB4 architecture with fine-tuning layers. A collection of 6862 photos with 11 annotated classes, including dew, fog/smog, frost, glazing, hail, lightning, rain, rainbow, rime, sandstorm, and snow, served as the basis for training and testing the model. The dataset was randomly divided into three different sets for training (80%), validation (10%), and testing (10%). Data augmentation was then employed on the training set to remove class imbalances. The proposed model's accuracy at predicting the weather was 92%, proving its effectiveness. The model adapted to the specific assignment of classifying the weather thanks to the fine-tuning layers, and the EfficientNetB4 architecture offered a solid and effective feature extractor. The model was trained using the Adam optimizer, which adjusted the rate of learning for each of the parameters separately and assisted the model to converge more quickly and consistently. The suggested model may benefit various applications, including transportation, agriculture, aviation, and emergency management, that call for precise and fast meteorological data. Additionally, it can enhance machine learning and digital image processing. Future studies may further investigate other architectures, hyperparameters, and regularization methods to boost the model's speed and resilience.